The 11th International Symposium on Wireless Personal Multimedia Communications (WPMC’08


Luca Chiaraviglio Politecnico di Torino Torino, Italy A BSTRACT The sensitiveness toward energy consumption problems is driving Telecommunications operators to optimize network equipment utilization. Since cellular systems are often dimensioned for peak hour traffic, during low traffic periods, such as night, many devices are underutilized but still, by being active, consume power. In this paper, we show that dynamic planning, consisting in reducing the number of active access devices when traffic is low, can achieve significant power saving. In our study, we consider three different UMTS scenarios with a simplified traffic model describing three classes of services, quality of service guarantees, link-budget, propagation and electromagnetic exposure constraints. I I NTRODUCTION Delia Ciullo Politecnico di Torino Torino, Italy Michela Meo Politecnico di Torino Torino, Italy II Marco Ajmone Marsan Politecnico di Torino Torino, Italy


II.A UMTS Radio Access Architecture UMTS, commonly referred to as 3G (for 3rd Generation Wireless Mobile Communication Technology), carries many types of traffic, from real-time circuit-switched to packet-switched services, and offers higher data rates and a wide range of telecommunications services, including videocall and Internet access. The UMTS architecture is typically composed by three interactive domains: User Equipments (UE), Terrestrial Radio Access Network (UTRAN), and Core Network (CN). The UTRAN provides connectivity between the UE and the CN. Basically, the UTRAN consists of two elements: the Node B, that is the base transceiver station, and the Radio Network Controller (RNC), that controls one or more Node B’s. While Node B and RNC can be co-located in the same device, typical implementations have a RNC located in a central office serving several Node B’s. As reported in [9], an individual UTRAN equipment consumes about 6 kW, including power amplifiers, digital signal processors, air-conditioning modules and feeders connecting the RNC to Node B. Considering only the Node B, its power consumption amounts to nearly 800 W [6] with power needed to transmit from the antennas usually in the range 1-40 W. The large number of equipments in the UTRAN makes the total power consumed at the access particularly significant: any reduction of power consumption at the UTRAN equipments can translate into a significant reduction of the overall consumption. In order to model Node B behavior, we use a traditional teletraffic model based on Markovian assumptions for the involved processes; the model allows us to evaluate cell performance in terms of blocking probability for a number of different service types. Moreover, we use a link-budget evaluation and a popular attenuation model to estimate the transmission power needed at the Node B’s. We focus on an urban scenario, in which cells can be assumed to have similar size and traffic is homogeneous. II.B Cellular Model We use a teletraffic model based on a multi-class M/M/N/0 queue, that was widely used in the literature for cellular systems [8]. The model focuses on a single cell in a cellular system, approximating the interaction between neighboring cells through a simplified description of the flow of handovers. There are K classes of service with different resource needs, a class i call requires an amount of bandwidth equal to Ci . The model relies on the following simple assumptions: i) users generate class i calls according to Poisson process with

Nowadays, energy consumption has become a key issue, from both environmental and economic side. ICT alone is responsible of a percentage which varies between 2% and 10% of the world power consumption [12] and this figure is expected to grow further in the future. Telecommunications, in particular, are greedy energy-user. For example, the energy needed by the major telephone operator in Italy is more than 2TWh a year, representing the 1% of the total national demand [11]. In the last years we have witnessed a consistent increase in the number of mobile users [5], and new services like videocall and video on demand have become very common in mobile networks. This trend has brought to the development of complex access infrastructures [10] with high energy consumption. Recently, telecommunication operators have become aware of the energy issue, and have begun to study new hardware solutions that can improve energy efficiency of the access network [9]. In this paper, we consider the possibility of a dynamic planning that, based on traffic intensity, reduces the number of active access devices when they are underutilized, such as during night periods. When some base transceiver stations are switched off, radio coverage and service provisioning are taken care of by the devices that remain active. The switching off of the access devices should be carefully decided so as to maintain quality of service guarantees and meet electromagnetic exposure constraints. In particular, we consider a UMTS radio access network and we take into account the possibility to turn off the Node B equipments. We claim that it is possible to apply these schemes to a real urban network, due to the fact that operators are providing new software features to control and even switch equipments off [4]. Considering the large number of access devices, the total energy saving for an operator can become huge.

ni .i . nK ) = K (2) s∈S where ρi = (λi + λh.e. that is represented by the parameter λh. during off-peak periods the system is probably overprovisioned and may waste a significant amount of power. The traffic model is extremely simple. Thus. equal to 12. We consider a set of cells with the same radius R and the same load. PM S . as reported in [3].C Link-Budget and Propagation Model The base station site configuration defines the maximum allowed path loss. µh is computed as. iv) class i service time is distributed according to a negative exponential pdf with mean 1/µi . n2 .16 π (s) with K i=1 Si = {(n1 . iii) no queuing is possible. After computing the allowed path loss. with rates. Class i blocking probability is given by the probability of the states in which an additional class i call cannot be accepted due to lack of available bandwidth: Pb. nK )| i=1 Ci ni ≤ CT } (1) where CT is the total UMTS maximum transfer rate. 1/µi . is derived by a fixed-point iterative procedure in such a way that. collected in the vector s = (n1 . since the Base Station transmission power.i = s∈Si Table 2: Link-Budget main parameters for the Downlink PBS AntennaGainBS EB /N0 P rocessingGain T otalN oise AntennaGainM S Sof tHODiversityGain SlowF adingM argin [dBm] [dB] [dB] [dB] [dB] [dB] [dB] [dB] Voice 33-37 15 4 25 -99 0 3 13. Thus. n2 . the main configuration parameters are shown in Table 1 for the Uplink (UL) and in Table 2 for the Downlink (DL). · · · .5 18 -99 2 3 13.5 14 -102 15 2 2 3 13.OFF S CHEMES In this section. that is the maximum power reduction of the signal between UE and Node B which still guarantees the communication. but adequate for a first estimation of the amount of energy saving that can be obtained with energy-aware planning. is given by.16 Data 24 2 1. · · · . n2 . the incoming and outgoing handover flows are equal. The most limiting parameter in the UL is the mobile station transmitting power. has to be shared among all users. The value of µh depends on user mobility. e. we consider K = 3 classes of service. our objective is to switch off some cells when load is low: we have to decide the number of cells to switch off . The cell is described by the number of active class i calls.16 Video 33-37 15 2. the set of all possible states. ii) incoming handovers for class i calls occur according to a Poisson process with parameter λh. The interaction between neighboring cells. i. v) the time spent by the user in the cell is distributed according to a negative exponential pdf with mean 1/µh . However. at peak traffic. The maximum path loss computation is based on the typical link-budget parameters. at steady-state.2Kbit/s. 64Kbit/s and 384Kbit/s or 144Kbit/s. i.g. we describe a procedure to verify the feasibility of a cell switching-off scheme and to compute the possible power consumption reduction. in our model.16 S = {s = (n1 . For more details see [10].. n2 . The DL direction limits the available capacity of the cell. nK ). Radio coverage and cell dimensioning are typically performed so as to satisfy quality of service constraints under peak traffic conditions. blocking probability for each class must be smaller than a target value (T) Pb . respectively. nK )|CT − Ci < Ci ni ≤ CT } (3) The average number of active class i calls is: E [ Ai ] = s∈S ni π (s) (4) The typical applications supported by UMTS are voice. are set according to typical values used in the literature [2]: 3 minutes for voice. K II.16 Video 24 2 2 18 -102 15 2 2 3 13. µh = V 4ln(2)R (5) III S WITCH . the maximum loss among UL and DL. we have considered an average speed V equal to 10m/s in the residential scenario and 5m/s in the office one. The UMTS maximum transfer rate is about 2Mbit/s.The 11th International Symposium on Wireless Personal Multimedia Communications (WPMC’08) rate λi . The mean call durations. videocall/videoconference and data transmission.16 Data 33-37 15 2 14 -99 2 3 13.. calls are blocked if the available bandwidth is not sufficient to satisfy an incoming request. PBS . the steady-state probabilities are: i ρn i n! i=1 i i ρn i n! i=1 i K π (s) = π (n1 . From well-known queueing theory results.. we use the well-known Walfish-Ikegami propagation model [1] to calculate the maximum cell radius Rmax . · · · . · · · .e. so that given the cell radius R. 5 minutes for videocall and 15 minutes for data connection.i . Table 1: Link-Budget main parameters for the Uplink PM S AntennaGainM S EB /N0 P rocessingGain T otalN oise AntennaGainBS DiversityGain LN A Sof tHODiversityGain SlowF adingM argin [dBm] [dB] [dB] [dB] [dB] [dB] [dB] [dB] [dB] [dB] Voice 21 0 5 25 -102 15 2 2 3 13. The state space.i )/(µi + µh ) is the class i load.

but in order to prove that significant energy saving are achievable. Every µ-cell has a radius R = 100m and total power (T) PBS = 2W. becomes R′ = 200m. PBS .i )/(µi + µh ). a new scheme with a smaller number of Coff cells should be assumed. The night zone extends from about 10 p. In this way the on cells coverage area doubles. If exposure limits are guaranteed. which represents the number of off cells for each on cell. The functions Λi (t) have the typical periodic night/day pattern. in other terms. IV S IMULATION R ESULTS IV. a simple sinusoidal behavior is enough. in order to increase the radius. traffic in t∗ is the peak traffic of the night configuration with Con on cells. cells could require additional transmission power.The 11th International Symposium on Wireless Personal Multimedia Communications (WPMC’08) and the load conditions under which cell switching off is possible. the longer the nigh zone is. the extremes of this interval corresponding to the peak hour traffic for the night scenario. the call generation rate doubles. First. Let the average number of active calls at time t∗ be denoted by E [Ai ](t∗ ). 4 reports the blocking probability perceived by each class of service during the two day period of analysis. to 7 a. and power con′ sumption reduction can be computed from the value of PBS and the extension of the night zone. the time varying class i call arrival rate is Λi (t) = αi Λ(t). Pconn = PBS E [Ai ](t∗ ) (7) Figure 1: The procedure adopted to verify a switching off scheme Figure 2: Switch off half of microcells with constraint (6). the load is ′ ′ ′ ′ ρ′ i = (λi + λh. with k depending on the geometry. and. the time period during which we can switch cells off while maintaining quality of service guarantees. the higher the power consumption reduction we can achieve. Fig. if either exposure limits cannot be guaranteed or no nigh zone is possible 1 Actually. respectively. the traffic is given by λi = Λi (t). Given the total base station power PBS . Let Coff and Con be the number of cells that. they have the simple sinusoidal shape that we assumed for night/day patterns. the scheme is feasible. In order to cover the area left by off cells. are switched off and that remain on. when Coff cells are switched off. At instant t. Assuming that every Node B covers two µ-cells.m. in the worst case. 1. The corresponding value of t.A Residential Scenario We considered a typical residential scenario. Otherwise. These decisions are critical and should take into account two aspects. the larger cell radius means also traffic load increase. with µh given by (5) substitut′ ′ ing R to R. 31 . Clearly. the cells that remain on must provide radio coverage over the whole area (including the portions that were covered by off cells). . Let Λ(t) be the time varying function of the new call generation rate. composed by µcells. real traffic patterns have more complex period behaviors. A possible configuration is represented in Fig. We call night zone. the and use it in the link-budget and the Walfish-Ikegami attenuation model to find the maximum cell radius Rmax . namely t∗ . in normal conditions. 3 for two consecutive days. The night zone is defined by the largest value of Λ(t)′ = (x + 1)Λ(t) that guar(T) antees that blocking probability is smaller than Pb for each service class.i ((x + 1)Λi (t∗ )) ≤ Pb ∀i (6) plays a role similar to the peak hour used for dimensioning the system. Blocking probability is always below the target.i derived from the handover flow balance iterative procedure. The quality of service target is Pb = 1%. the traffic in on cells becomes λ′ i = (x + 1)λi . we define the power per connection as. correspondingly. x is the scaling factor Coff /Con .. and electromagnetic exposure limits as described in [7] must be verified. and λh. during low traffic periods (say. Otherwise.m. If Rmax > R′ the switching off scheme can be implemented without increasing the power PBS . such that (T) Pb. The procedure we proposed is discussed below and sketched in Fig. such as those reported in Fig. under which quality of service constraints must be guaranteed. a new value for the base ′ station power must be computed. which we partition into service classes according to some constants αi . on cells radius must increase from R to R′ = kR. nights). and the new radius R′ . 2: we set x = 1 and switch off one µ-cell every two. half of Node B’s can be switched off. Second. The functions Λi (t) are reported in Fig.

the maximum cell radius is always larger than 200 m. The maximum cell radius computed from the propagation and linkbudget models under traffic Λi (t)′ is reported in Fig.001 Pb 0. The first case is similar to the previous one. During the night zone.03 0. Due to the lack of space.06 0.05 Night zone 0.5% of power consumption in a day. In the second case. Every Node B can be switched off for about 9 hours. 5. an on cell should cover the area of 1.07 0. and three cells out of five are switched off. 0.04 λ 0. a Node B controls three cells. Figure 4: Blocking probability (Residential scenario) 12 10 8 E[call] 6 4 2 λ 0 8:00 16:00 2λ 24:00 8:00 Time [h] λ 16:00 2λ 24:00 8:00 VOICE VIDEO DATA Table 3: Results for the Office Scenario x λ′ R R′ PBS P ′ BS Tnight Saving (N odeB ) Case 1 1 2λ 100 200 2 2 12 50 Case 2 3/2 2. Similar behavior is shown by the average number of calls in Fig. thus.B Office scenario In a typical urban area. with the smaller values of maximum radius. no base station transmitted power increase is needed.01 0. the call generation rate Λ(t) becomes negligible during night.1 VOICE VIDEO DATA 0. there are many office zones.5λ 100 400 2 5 6 25 Figure 5: Average number of active calls (Residential scenario) [m] [m] [W] [W] [h] [%] .01 0 8:00 Night zone VOICE VIDEO DATA Rmax [m] 300 280 260 240 220 200 180 160 8:00 16:00 24:00 8:00 Time [h] 16:00 24:00 8:00 16:00 24:00 8:00 Time [h] 16:00 24:00 8:00 VOICE VIDEO DATA Figure 3: Call generation rate versus time in a µ-cell (Residential scenario) Figure 6: Maximum radius achievable when half of the cells are off (Residential scenario) maximum is achieved at peak day and night hours. we report in Table 3 some results obtained in two situations that differ for the geometry of the cellular planning. In particular.02 0. in the worst case.5 off cells. The potential saving is 800W × 9h = 7. only two cells are in charge of covering the area of the three cells that are switched off so that. and. saving 37.2KW h.The 11th International Symposium on Wireless Personal Multimedia Communications (WPMC’08) 320 0.0001 1e-05 1e-06 1e-07 8:00 λ 16:00 2λ 24:00 8:00 Time [h] λ 16:00 2λ 24:00 8:00 IV. The main difference with respect to residential areas is in the load night/day pattern. 6: data traffic is the limiting case.

Lempiinen and M. Manninen. 2008. while in order to guarantee continuous coverage.04 Rmax [m] 0. 2001.ptinovacao.33dBm. [4] Vodafone using Ericsson’s new power-saving base station feature. Chapman and Hall CRC.cellular-news.03 [10] J.htm. we showed that it is possible to switch off some cells and Node B’s in urban areas during low-traffic periods. We analyzed three kinds of scenario: residential. Link-Budget limits give the maximum radius Rmax versus Pconn shown in Fig. IEEE Transactions on Vehicular Technology. [2] Characterisation parameters for enhanced UMTS services and Carciofi. Evaluation of exposure levels generated by cellular systems: methodology and results. [7] M.The 11th International Symposium on Wireless Personal Multimedia Communications (WPMC’08) 0. Seacorn Project Public Deliverables. 2005. [12] Global Action An inefficient truth. http://seacorn. 7. November 2002. [5] GSM family subscriptions reach 3 billion http://www. In all the scenarios it is possible to reduce power consumption.ppt. http://www.lx. 2008. Seacorn Project Public Deliverables.motorola.01 0. .3g. Energy and power conversion: A telecommunication hardware vendor’s perspective.05 0.pdf. 51(6):1322–1329. C. Power Electronics Industry Group. In particular. New York. when the limit is given by the mobile station transmission that corresponds to transmission power PBS of about 3. Pileri. Hodes. USA. 50% savings can be achieved in some˜1. so as to turn off the Node B’s that control them. Global Action Plan Report. Blocking probability is reported in Fig. 2002. Handbook on Algorithms for Wireless Networking and Mobile Computing. The switching off scheme consists in switching off 7 µ-cells during night. Barbiroli. Boukerche. 2007. Radio Interface System Planning for GSM/GPRS/UMTS. 8: during night the call generation rate at the umbrella cell is seven times the one of the single cell. 2007. 2002.4W. worldwide.e.02 http://www. The curves are flat when the constraint is the UL. http://www.01 0 8:00 λ VOICE VIDEO DATA 600 550 500 450 400 350 300 250 200 150 15 16:00 24:00 8:00 Time [h] 16:00 24:00 8:00 20 25 Ptxconn [dbm] 30 35 VOICE VIDEO DATA Figure 7: λ variation for a single µ-cell (Hierarchical scenario) 1 [9] M. i.C Hierarchical scenario Some systems have a hierarchical cellular structure in which umbrella cells are used to cover shadowed regions of smaller cells and fill gaps in coverage between those cells or to manage overflow traffic. http://seacorn. 2007. MA. R EFERENCES [1] Cost 231final office and hierarchical. http://www. http://www. we showed that a large amount of energy can be saved if a careful dynamic radio coverage planning is used instead of a static pileri. In our scenario an umbrella cell covers seven µ-cells. Energy and communication: engine of the human progress.ega. 2007. With radius 265 m.0001 1e-06 1e-08 Pb 1e-10 1e-12 1e-14 VOICE 1e-16 VIDEO DATA λ 1e-18 8:00 16:00 Figure 9: Maximum radius variation versus average power per connection (Hierarchical scenario) V C ONCLUSIONS AND FUTURE WORK 7λ 24:00 λ 8:00 16:00 Time [h] 7λ 24:00 8:00 In this paper. 9. [6] Node B datasheets.06 0. The radius of the circular umbrella cell is about 265m. the call generation rate is reported in Fig. the saving is double if two Node B’s can be switched off. of 100m radius and PBS = 2W. [8] A.php. Kluwer Academic Publishers. [11] S. Intelec Opening Degli-Esposti.globalactionplan. the umbrella cell is always on. All the 7 µ-cells covered by the umbrella cell can be switched off for about 4 hours. A 17% saving of power consumption in a day can be achieved if a single Node B controls the 7 µ-cells. USA. while still guaranteeing quality of service constraints in terms of blocking probability and electromagnetic exposure limits. Figure 8: Blocking probability (Hierarchical scenario) IV.. [3] Classification of mobile multimedia services. the umbrella cell needs a minimum Pconn = 21. and V.

Sign up to vote on this title
UsefulNot useful